Air Canada and the Chatbot: the AI Lesson That Cost Dearly

Air Canada returned to the top of search results in 2026 — record revenue of US$5.8 billion in the first quarter, new cabins on the A321XLR and B787-10, and more flights connecting Brazil and Canada. But the most useful story for those using AI in customer service is not in the balance sheets: it's a case where the company's chatbot cost dearly. Understand the lesson.

TL;DR

• Air Canada is in the spotlight for strong numbers in 2026, but its most instructive episode for businesses is the lawsuit over its website chatbot.
• In 2024, a Canadian court held Air Canada liable for a wrong answer from its chatbot and ordered it to compensate the customer.
• The defence that "the chatbot is a separate entity" was rejected: the company is responsible for everything its AI says.
• The lesson applies to any business using a bot or AI agent — including on WhatsApp.
• Content governance, logs, and human review are no longer a luxury but a requirement.

Why Air Canada is trending in 2026

Interest in "Air Canada" has surged this first half for legitimate business reasons. The company started 2026 with record operating revenue of US$5.8 billion in the first quarter, up more than 11% over the same period in 2025 — although it suspended annual projections due to fuel volatility.

In Brazil, the topic gained traction because the bilateral relationship with Canada is expanding, with increased international mobility and air connectivity. Add to that the launch of new cabins on the A321XLR and Boeing 787-10 jets, and you have a brand that never leaves the news.

However, for those working with technology and customer service, there is a chapter of Air Canada that teaches more than any results release. It involves artificial intelligence, a grieving customer, and a court.

The case that changed everything: the Air Canada chatbot

In November 2022, Jake Moffatt needed to fly urgently after his grandmother's death. Before buying the ticket, he chatted with the Air Canada website chatbot to understand the bereavement fares. The bot informed him that he could buy the ticket at full price and request a retroactive refund of the difference within 90 days.

Moffatt followed the guidance. When he requested the refund, Air Canada denied it: the actual policy, published on another page of the same website, stated that bereavement fares do not apply retroactively. In other words, the chatbot contradicted the company's official documentation.

What the chatbot promised (and was wrong)

The key point is that the wrong information did not come from an unprepared agent. It came from an automated system, on the official channel, which the customer had every reason to consider reliable. Moffatt took the case to the British Columbia Civil Resolution Tribunal.

The defence the tribunal rejected

Here is the part that deserves a print on the wall of any product team. Air Canada argued, in essence, that the chatbot was a separate legal entity, responsible for its own information. The tribunal did not buy it. In the decision Moffatt v. Air Canada (2024 BCCRT 149), arbitrator Christopher Rivers was direct: the chatbot is part of Air Canada's website, and the company is responsible for all information appearing there — whether from a static page or a bot.

Result: Air Canada was ordered to pay CA$812 to Moffatt, the difference between the bereavement fare and the full price he paid. The amount is small. The precedent is not.

What the decision means for companies using AI

The legal thesis is simple and powerful: you cannot outsource responsibility to your own AI. If the channel is yours, the bot's speech is yours. For Brazilian companies, the reasoning fits directly into the Consumer Protection Code, which treats information provided to consumers as binding and holds suppliers liable for misleading advertising and data.

In practice, this debunks three common illusions:

1. "The bot is experimental, so it doesn't count." If it is live serving customers, it counts as an official company statement.
2. "It's the AI provider's fault." For the consumer, the brand they hired is responsible.
3. "A footer disclaimer protects me." The tribunal understood that the customer acted reasonably in trusting the Air Canada chatbot — generic disclaimers do not override a specific and incorrect instruction.

Air Canada is not an exception: why this applies to your business

At Agathas Web, I have been implementing AI agents for customer service long enough to state: Air Canada's mistake was not using AI, it was using AI without governance. The chatbot had access to answers, but was not anchored in the company's source of truth.

I see the same risk in small businesses that plug a generic bot into their company WhatsApp and disappear. The bot "hallucinates" a promotion that does not exist, promises an impossible deadline, or invents a return policy — and the owner only finds out when the customer complains. The difference between Air Canada and the corner bakery is the scale of the loss, not the nature of the error.

If you are evaluating how to bring AI to your main channel, it is worth first understanding the difference between the WhatsApp Business App and the Official API, because the level of control over what the bot responds changes completely between the two.

The 5 mistakes that led Air Canada to court

The case is a lesson in what not to do. I summarised the mistakes and the corresponding antidote:

Air Canada's Mistake	What was missing	Best practices
Bot responded without checking official policy	Anchoring in the source of truth	Connect the agent to the updated knowledge base
Bot information contradicted the website	Consistency across channels	A single source feeding bot and pages
No review of sensitive responses	Human curation	Escalate to human for critical topics
Defence of "separate entity"	Clear accountability	Assume the bot as the brand's official voice
Lack of robust audit trail	Traceability	Log every conversation for review

None of these points require cutting-edge technology. They require process. And process is exactly what is often missing when AI is implemented "on the fly".

How to safeguard your AI-powered customer service: best practices

The good news is that the Air Canada case left a clear prevention roadmap. I treat these pillars as non-negotiable in any project.

Content governance

The agent should never "invent" policy. It responds from a controlled knowledge base — prices, deadlines, rules — that reflects the website and contracts. When the information is not in the base, the correct response is "I'll transfer you to a human", not an educated guess.

Logs and traceability

Every conversation needs to be recorded. Without logs, you cannot audit what the bot said, train improvements, or defend against a complaint. In the Air Canada case, it was precisely the chatbot history that sealed the liability.

Human review on sensitive topics

Bereavement, refunds, health, legal, cancellations. Whenever the topic has emotional or financial weight, the flow should offer a human. Automate the volume; supervise the critical. This is how to scale service without repeating Air Canada's mistake.

Those who want to understand how well-designed AI agents operate in this balance can check what Gemini Spark changes for businesses — the logic of a context-connected agent goes against the generic bot that brought down the airline.

Generic chatbot vs. well-implemented AI agent

It is worth separating two worlds that many people confuse. The generic chatbot is the one with pre-set answers or loose AI, without connection to the company's data — exactly the profile that failed at Air Canada. A modern AI agent, on the other hand, is orchestrated: it consults the base, follows business rules, knows when it does not know, and escalates to a human.

The practical difference appears in the cost of the error. A well-implemented agent will not promise a retroactive refund if the policy says otherwise, because the policy is the source it consults. This is the kind of architecture I advocate when it comes to putting AI on the highest-volume channel for companies, WhatsApp — and which I explain in why charging per employee in customer service failed.

The future: AI in customer service after the Air Canada case

The Air Canada precedent is already cited in regulatory discussions worldwide about liability for autonomous systems. The trend is clear: the more autonomous the AI, the more explicit the company's chain of responsibility needs to be.

This does not slow adoption — quite the opposite. Companies that treat AI seriously gain a double advantage: they serve faster and with less risk. Those that treat the bot as a gimmick end up, sooner or later, writing their own "Air Canada" chapter.

The bar has changed. Reliability is no longer a differentiator but a prerequisite for putting an agent to talk to your customer.

And in Brazil? What the CDC says about AI in customer service

The precedent is Canadian, but the reasoning reaches Brazil without friction. The Consumer Protection Code establishes that offers and information provided to consumers are binding and form part of the contract (articles 30 and 48), and that misleading advertising creates supplier liability. It does not matter whether the speaker was a salesperson, a page, or a bot: if the message came from your channel, it binds you.

There is a local aggravating factor. The CDC works with strict liability of the supplier — meaning the consumer does not need to prove fault to be compensated, only the damage and the connection to the service. In practice, a Brazilian company that repeated Air Canada's mistake would have, at minimum, the same outcome, possibly with moral damages added to the material ones.

The General Data Protection Law (LGPD) adds another layer when the bot handles personal data and automated decisions: the data subject has the right to review and explanations. In other words, having a human in the loop and auditable logs is not only good customer service practice but also legal protection.

Conclusion: the Air Canada lesson for your company

Air Canada is experiencing a strong 2026 in numbers, but its most useful legacy for the rest of us is a CA$812 ruling. It teaches that AI in customer service is powerful and holds those who use it accountable. You cannot automate the conversation and de-automate responsibility.

If you intend to put an AI agent to serve your customers, start with the basics that Air Canada ignored: anchor the bot to the source of truth, log everything, and keep a human in the loop for what matters. Want to review your AI-powered service flow before scaling? Talk to us — it is better to design governance now than to discover its cost in court.

Sources: CBC News{target="_blank"}, legal analysis by McCarthy Tétrault{target="_blank"} on Moffatt v. Air Canada (2024 BCCRT 149), and 2026 results data released by aviation press.